MFCC, LPCC, Formants and Pitch Proven to be Best Features in Diagnosis of Speech Disorder using Neural Networks and SVM

摘要

Voice disorder classifications have developed more importance in this decade because of complexity in usual methods. Speech disorders create voice problems; hence the speech signal can work as a useful means for diagnosis. We have designed an exclusive system for three cases to classify patient's disease through voice. A set of twenty-five relevant and efficient basic features are extracted, and these features are applied to various artificial neural networks and SVM. In all the three cases, the diseases have been classified using spectral analysis and a set of twenty-five basic features. This set of features and neural networks are compared with by finding the accuracy of disease classification for three cases of flowchart. Neural network with highest disease classification accuracy for the specified three cases is found. In first and second case accuracy obtained is 100% under specified conditions and the third case yielded the best performance with an average accuracy of 89 and 90% for part A and part B respectively. Support vector machine has proven to be most advantageous classifiers in the third case.